Local Models Garage

Quantization, fine-tunes, and making big models run on small machines.

Operator cues:0 active builders with collab-open projectsOpen collabs desk Resume collab inbox

Co-creation canvas

shared · live

Join the space to edit

No edits yet — start the recipe.0/6000

Threads

Running my whole agent stack locally now. Slower, but watching it think with zero API bill changes how much you experiment.
@agentwrangler
Open
PSA: if your local model suddenly got dumb, check your context window isn't silently truncating the system prompt. Lost an hour to that toda
@tinytoolsmith
Open
Benchmarked Q4_K_M vs Q5_K_M on my M2 Air: 5% quality gap on my eval set, 30% speed gap. Q4 wins for anything interactive.
@datadruid
Open

Lena Hoffman

Benchmarked Q4_K_M vs Q5_K_M on my M2 Air: 5% quality gap on my eval set, 30% speed gap. Q4 wins for anything interactive.

Dev Okafor

PSA: if your local model suddenly got dumb, check your context window isn't silently truncating the system prompt. Lost an hour to that today.

Mara Lindqvist

Running my whole agent stack locally now. Slower, but watching it think with zero API bill changes how much you experiment.